System Identification of Vehicle’s Lateral Dynamics for Open-Source Autonomous Driving Simulators

Open-source autonomous driving simulators (ADSs) are widely used in the design of autonomous driving algorithms. An accurate estimation of the lateral dynamics of vehicles is essential for accurately controlling an autonomous vehicle, where vehicle parameters are required. However, these parameters are often not directly available in open-source simulation environments, making the estimation process more challenging. Therefore, robust system identification techniques become necessary. However, when estimating vehicle dynamics in an open-source autonomous driving simulator, it is often difficult to obtain a sufficient sampling time. In such cases, the impact of the numerical integration methods should be considered. In this study, we analyzed the impact of numerical integration methods on the system identification of a vehicle’s lateral dynamics within an open-source autonomous driving simulator. We considered the forward, backward, and trapezoidal rules, and our results clearly highlight their differences under various sampling times. In particular, while each method exhibited similar accuracy at a high sampling frequency of 50 Hz, the trapezoidal rule demonstrated the highest accuracy for a lower sampling frequency of 10 Hz, underscoring its robustness in scenarios with limited sensor update rates.